University of Michigan researchers build prototype of lie-detecting software on the basis of real world data
11 Dec 2015
Researchers from the University of Michigan have built a prototype of a lie-detecting software on the basis of real-world data.
The prototype factored in both the words and gestures and unlike a polygraph, it does not need to touch the subject in order to work.
The detector was found to be 75 per cent accurate in identifying who was being deceptive (as defined by trial outcomes), as against humans' scores of just above 50 per cent.
With the software, the researchers said they had identified several lies.
According to the authors, lying individuals moved their hands more, and tried to sound more certain. Also, they looked their questioners in the eye a bit more often than those presumed to be telling the truth, among other behaviours.
According to Rada Mihalcea, professor of computer science and engineering who leads the project, there were clues that humans gave naturally when they were being deceptive, but people did not pay close enough attention to pick them up.
According to commentators, the system might one day emerge as a helpful tool for security agents, juries and even mental health professionals.
In the development of the software researchers used machine-learning techniques to train it on a set of 120 video films from media coverage of real trials.
The team also used some video footage from The Innocence Project website, a non-profit legal organisation committed to exonerating wrongly convicted people.
According to Mihalcea, the 'real world' aspect of their work was one of the main ways it was different.
The research team counted and listed the gestures used by the speakers in the videos, using a standard coding scheme for interpersonal interactions that scored nine different movements of the hands, mouth, brow, eyes and head.
The data was then fed into their system and they waited for it to sort the videos. With input from both the speaker's gestures and words it was 75 per cent accurate in determining who was lying.
